Abstract: Aiming at the time-varying observation noise problem faced by UAV (unmanned aerial vehicle) optoelectronic systems in multi-target passive localization, a joint adaptive extended Kalman filter (JAEKF) algorithm is proposed to realize passive geo-localization of ground targets. The algorithm adopts a three-level hierarchical processing architecture: the classical extended Kalman filter framework is constructed at level 1 to achieve the baseline estimation of the target geographic coordinates, and effectively suppress the static observation noise interference; the dynamic residual covariance matrices is analyzed in real time at level 2 to adaptively adjust the weight distribution of the predicted and measured values in the observation model; and the adjustment mechanism for the time-varying forgetting factor is designed at level 3 to make the predicted values of the observation covariance adaptively adapt to the changes of UAV flight attitude, forming a double adaptive compensation system. Simulation and real flight experiments show that the average positioning accuracy of the proposed algorithm for multiple targets is 14.69 m, and the error range is 1.87~5.21 m in the condition of time-varying noise, indicating that the algorithm has the characteristics of high accuracy, strong stability and good real-time performance.